Thanks for the example. I'll take a look.
To confirm, the algorithm for the exact/symbolic case did change in version 12.3. But the intent was not to degrade performance.
I would guess this is not in the QRDecomposition part but rather in some simplification algorithm it uses internally to simplify expressions. Do you need the result to be symbolic? Because you're pushing the limits I think with doing QR decomposition with quite big symbolic matrices…
EDIT: I was wrong; but good for Daniel to look into :-)
Hi, Thanks very much for your quick reply, I wish the result to be symbolic, it is constructed from some structure constants of SU(3) group, it is expected to be nice symbolic forms